[Silthus]

Merge gate: merge only after #131567 is deployed in both regions, and cyclotron_email_attachments has zero rows in both regions on the day it merges. The query is in the ticket's proof comment. The drop deletes data, so it can't be undone.

Problem

The side table cyclotron_email_attachments held the file bytes of transactional email in the cyclotron job database.
Email attachments are moving to object storage, and #131567 removed every reader and writer of the table.
The empty table stays in the shared queue database until something drops it.

Refs PostHog/michaelr-context#74

Caution

🔴 Merge risk: high · one-way door: the table and any rows in it are deleted

• If this merges before feat(cdp): remove inline email attachments and their side table #131567 runs everywhere, pre-removal CDP pods log a caught error on each janitor sweep. Sends and job cleanup carry on, in both regions.
• Switch-off: none. The migration runs on deploy, then fix forward.

Rollout

• Stages: feat(cdp): remove inline email attachments and their side table #131567 deployed in US and EU, then the zero-row check in both regions, then merge. The sqlx migrate deploy applies the drop.
• Watch: the sqlx-migrate deploy for cyclotron-node, and CDP janitor errors.

Changes

• A new cyclotron-node migration drops cyclotron_email_attachments. Its primary key and two indexes go with it.
• Nothing a user sees changes. Code sends behave as on master.

How did you test this code?

All runs were on a dev stack at this PR's head. The evidence and the merge-day query are in the ticket's proof comment.

• Red/green: master's migrations leave the table and its three indexes in place. This head's migrations remove them. A second run is a no-op, and cyclotron_jobs is untouched.
• Live run: the real migration runner applied the drop. A code send through Django, cdp-api, the email worker and maildev was delivered. Several janitor cycles ran with no error about the table. A send with attachments is still refused with field_not_accepted.
• Cross-version: Node from before feat(cdp): remove inline email attachments and their side table #131567 still delivers on the dropped schema. Its janitor logs a caught "relation does not exist" on each sweep and keeps cleaning up jobs. The merge gate prevents this.
• Shadow diff: schema dumps of master and this head differ only by the table, its key and its indexes. Three send scenarios give the same results with the table present and with it dropped.

Test rationale: none added. A test that asserts a table is gone guards nothing once the migration has run. The migration runner and the schema dumps above prove the drop.

Release status

• No feature flag controls this change

Docs update

• None.

🤖 Agent context

Autonomy: Human-driven (agent-assisted)

Agent: Claude Code, Claude Opus 5.5. Reviews: GPT-6.1 Sol.

Skills: /writing-pr-descriptions, /writing-tests, /writing-code-comments

Question	Answer	Who	Where
Drop the side table once its code is gone	A separate migration, merged after the removal is live in both regions	👤 Human	PostHog/michaelr-context#74
Build ahead of the merge gate	Build and review now; the gate holds the merge	👤 Human	PostHog/michaelr-context#74
Lock timeout on the drop	None: the table has no readers once the gate holds, so the lock waits on nothing	🤖 Agent	PostHog/michaelr-context#74

Involvement: design pass, cross-model review, PostHog Review, qa-swarm; live proofs captured.

More context for AI agents

• Session: Claude Code in T3, stack worker for PostHog/michaelr-context#74
• Decisions: no .down.sql: this folder has none, and the fix is forward
• Read closely: rust/cyclotron-node-migrations/20261109000001_drop_email_attachments.sql

🤖 Generated with Claude Code

[github-actions]

🤖 CI report

✅ Trunk lane — non-backend lane

This PR is assigned to the non-backend lane. It does not run backend Python tests and may merge in parallel with PRs in other lanes.

[posthog]

PostHog Review · Standard · reviewed 19d4e1d: no findings.

[parameterai]

Risk: No findings

A single sqlx migration drops the now-unused cyclotron_email_attachments table from the cyclotron job database; no application code changes. I verified no code in the repo references the table anymore — the only matches are the create and drop migrations themselves — so nothing can break at runtime from the drop, and the statement is static SQL with no injection surface.

_{Sentinel reviewed 19d4e1d · Review settings}

[veria-ai]

PR overview

No security issues were found in this pull request.

Reviews

• 19d4e1d — No new issues

[stamphog]

Not approved yet — waiting on the conditions below.

Re-add the stamphog label to request another review once you have addressed this.

stamphog can't approve this pull request automatically. The deny-list gate refused it because the added file rust/cyclotron-node-migrations/20261109000001_drop_email_attachments.sql is a database migration, and the tier gate also classified the change as T2-never, a category that always needs a human reviewer. The size gate passed, so the size isn't the issue.

Next, ask a human reviewer to look at it, ideally someone who owns the cyclotron database. The merge-gate conditions in the description (the dependent change deployed in both regions and zero rows confirmed) are worth restating in that request.

• posthog[bot] reviewed the current head.

Gate mechanics and policy version

Gate		Result
prerequisites	✓	all clear
deny-list	✗	matches: migrations
size	✓	1L, 1F substantive — within ceiling
tier	✗	classified as T2-never: T2-never (1L, 1F, single-area, feat)
stamphog 2.7.0		.stamphog/policy.yml @ unknown · reviewed head 19d4e1d

[Silthus]

/trunk merge

[trunk-io]

😎 Merged successfully - details.

[Silthus]

/trunk merge

[deployment-status-posthog]

Deploy status

Environment	Status	Deployed At	Workflow
dev	✅ Deployed	2026-11-09 12:17 UTC	Run
prod-us	✅ Deployed	2026-11-09 12:57 UTC	Run
prod-eu	✅ Deployed	2026-11-09 13:23 UTC	Run